IMPRINT

The identity preservation engine for humans in the age of AI.

imprint.houseofnamus.com

AI does not replace you all at once. It replaces you one delegated decision at a time — a paragraph you did not write, a judgement call you did not make, a problem you did not sit with. The erosion is real but invisible, because there is no "before" to compare yourself against.

IMPRINT captures that "before" and keeps measuring the distance from it.

1 — Baseline capture. During onboarding you answer a set of prompts drawn from four universal modules (opinion & belief, decision under pressure, memory & recall, emotional fingerprint) plus modules specific to your profession cluster — language & voice, visual & creative, technical & analytical, human & social, leadership & strategy, or life & personal. Responses can be text, voice, or file upload.

2 — Metric extraction. Each response is reduced to four measurable signals:

3 — Calibration. On a recurring cadence you answer the same modules again. Each response is compared against its own baseline, per module.

4 — Drift Score. Four signals are combined into a weighted composite on a 0–100 scale, where higher means further from yourself:

Baseline divergence carries the heaviest weight because language degrades earliest and most visibly under delegation. The dashboard presents each signal inverted — as the quality drift erodes — so a fuller bar always reads as better.

The score is directionless by design — it measures distance from your baseline, not improvement or decline against anyone else.

• Drift — your score over time, with the contributing signals broken out so you can see which dimension moved.
• Calibration — the recurring re-test that produces each new score.
• Skill Vault — the skills you are deliberately protecting, each with a strength value and generated practice challenges.

• The Mirror — a Socratic reflection surface built on gpt-4o. It is constrained to ask questions and nothing else: it will not answer, advise, recommend, or write for you. It detects when you are asking it to decide something and redirects the question back at you, tracking how often that happens.
• The Forge — distraction-free composition with drift signals captured as you write.
• Journal, Beliefs, Time Capsule — longitudinal records of what you thought and when.

• Credential — a public, shareable page and generated badge image proving your baseline and current standing.
• Gallery, Profile, Settings, including an AI Reduction Protocol you can commit to for a fixed window.

• Circles — small accountability groups with check-ins.
• Mentors — request or offer mentorship.
• Leaderboard — ranked by imprint score.

Alongside the product, IMPRINT publishes what it is doing and why:

The scoring page exists because a score you cannot inspect asks for trust it has not earned. If the algorithm in app/api/calibration/complete/route.ts changes, /methodology is wrong and must be updated with it.

• Server-first. Dashboard pages are React Server Components that read from Supabase directly; client components receive data as props. Mutations go through route handlers under app/api/.
• Auth. Supabase SSR cookies, refreshed in middleware.ts. /dashboard/* and /onboarding/* redirect unauthenticated visitors to sign-in.
• Authorisation is in the database, not the app. Every table has RLS enabled and policies scoped to auth.uid(). Route handlers use the caller's session — the service role key is never used to bypass a policy.
• Prompt safety. The Mirror reads the user's cluster and baseline from the database rather than the request body, so a client cannot fabricate the baseline it is being measured against. The free-text session context is sanitised before it reaches the system prompt, and requests are rate-limited per user because every message is a billed model call.

Prerequisites: Node 18+, a Supabase project, an OpenAI API key.

git clone https://github.com/Sumandebnath943/imprint.git
cd imprint
npm install

Copy the example env file and fill it in:

cp .env.local.example .env.local

Apply the migrations in order from supabase/migrations/ via the Supabase SQL editor or CLI:

supabase db push

Then:

npm run dev

Other scripts:

npm run build          # production build
npm run lint
npm run seed:demo      # local demo data
npm run indexnow       # push the sitemap to Bing, Yandex, Naver, Seznam, Yep

indexnow is for after publishing, not for a timer — resubmitting unchanged URLs is what the protocol's 429 response exists to punish. Google does not participate in IndexNow; discovery there runs through Search Console and the sitemap.

app/
  (public)/       landing, about, methodology, drift-score, research, faq,
                  glossary, notes, for/[audience], courses, contact, legal,
                  public credential pages
  llms.txt/       generated agent-facing site map
  llms-full.txt/  generated full-text corpus
  (auth)/         sign in, sign up, password reset
  (onboarding)/   baseline capture flow
  (dashboard)/    the product
  api/            route handlers
components/       one directory per feature area
content/
  notes/          articles as MDX, read at build time
lib/
  beacon/         visitor beacon: geo, bot scoring, Telegram, crawler + probe ids
  seo/            entity constants, JSON-LD builders, route registry
  content/        glossary, FAQ, research index, cluster and note loaders
  supabase/       browser, server and middleware clients
  utils/          formatting, profession clusters and baseline modules
  validations/    zod schemas
  api/            shared route-handler helpers
scripts/
  indexnow.mjs    push the sitemap to IndexNow participants
supabase/
  migrations/     schema and RLS policies, applied in order
types/            shared domain types

Honest notes on where the MVP stops:

• Drift metrics are lexical (type–token ratio, sentence length, latency). They are deliberately transparent and cheap to compute, but they are a proxy for reasoning depth rather than a direct measure of it. Semantic and embedding- based signals are the natural next step.
• Courses are presented as upcoming with a waitlist; no course content ships in this build. /courses says so on the page rather than only here.
• A small number of credential statistics are placeholders pending the full relational rollup.
• The per-user rate limiter is in-process, so on serverless it is per-instance rather than global. Adequate for the current scale, and the seam for a shared store is isolated in lib/api/rate-limit.ts.
• The visitor beacon records IP, city-level location, interaction timings and — when signed in — the account, and sends them to a private Telegram chat. This is disclosed in the privacy policy, which carries a one-click opt-out (?notrack=1). Do Not Track is reported in the alert but does not itself suppress logging.
• Open Graph card images render in a fallback font. No font is passed to the image renderer, so every bold weight in components/seo/OgCard.tsx is currently inert and the cards are not in the brand typeface.
• Landing-page quotes in the friends-and-family section are placeholders. They are labelled as pre-launch impressions and carry no invented metrics, but they need replacing with real attributed words before launch. This also gates Review structured data — none ships anywhere on the site while they stand.

This README is the public documentation for the repository. The scoring method is published for readers rather than developers at /methodology, including a limitations section covering the length-sensitivity of type–token ratio, the directionless divergence measure, and the fact that three of the four signals measure engagement with IMPRINT itself.

Internal working documents — the project bible, the developer and portfolio handoffs, and the beacon write-up — are kept out of this repository under /docs/, which is gitignored.

MIT — see LICENSE.